Generally, it has proven difficult to reliably and efficiently load balance the demand for access to resources, e.g., a web-based application, email and streamed multimedia data, on a wide area network (WAN). One prior art attempt employed a look up table for storing a relationship mapping between a client's ip address and the ip address of the actual server that provided access to the resources for a domain name/ip address request. This table was usually held in the memory of a server array controller that managed several node servers that could provide access to the resources associated with the client's request. Typically, the server array controller would employ a load balancing technique to select and map the ip address of one of the managed node servers to the client's actual ip address and store this mapped relationship with a time stamp in the table. In this way, when a client repeated a request before the expiration of the time stamp, the controller would use the mapping stored in the table to automatically connect the client to the previously selected (load balanced) node server.
Additionally, if the time stamp had expired, the server array controller would again perform the load balancing technique to select one of the managed node servers to provide the actual access to the resources associated with the request. Each time the load balancing technique was performed, the controller would update the table to include a new time stamp and a new mapping of the client's unique ip address to the currently selected node server's ip address.
For a relatively small number of client requests, the above described prior art solution could reduce the demand on server array controller resources because the controller did not always have to perform a load balancing technique for each client request that occurred before the expiration of the time stamp. Instead, the controller only performed the load balancing technique for a new client request when the time stamp for a previous client request was expired. However, since all of the table entries had to be kept in the memory of the server array controller to be used effectively, the available controller resources for load balancing and managing several node servers decreased in proportion to an increase in the number of client requests. To ensure that table entries were not lost when the server array controller lost power or was rebooted, a copy of the table would be stored on a secondary storage medium. Also, under heavy load conditions, the secondary storage medium was often not fast enough to store the copy of table entries before the server array controller shut down.
Another significant problem with the prior art approach was that the client's ip address was not always unique. Although some clients might have their own unique ip address, many others used random virtual client ip addresses provided by a large Internet Service Provider (ISP), e.g., the America On-Line Corporation, to connect to the Internet. Since only a portion of a large ISP's clients are typically connected at any one time, a large ISP usually employs a proxy cache to randomly assign a relatively small number of virtual client ip addresses to the currently “on-line” (customers) clients. Typically, a proxy cache will assign one of the virtual client ip addresses to a client on a first available basis each time the client connects to the ISP and starts a session on the Internet. From the discussion above, it is apparent that when a client used a large ISP to connect to a WAN such as the Internet, the prior art did not provide an effective method for persistently mapping a client's relationship to the server that was selected to provide access to resources associated with a request.
Therefore, it is desirable to provide a method and system for automatically providing a persistent mapping of a previously selected destination for a domain name/ip address request. Preferably, the present invention employs a Cookie in a Hyper Text Transport Protocol (HTTP) data stream to identify a relationship between a previously selected destination and a client's HTTP request. The present invention overcomes many of the limitations of the prior art caused by the direct mapping of an actual destination ip address to a client's ip address.